Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Lmcli v0.7.0, a general-purpose LLM harness written in Go, has been released with smooth performance at 1M+ token context and features including composable agents, custom tools, sandboxed execution, and branching convers…
XAI launched the Grok Build Plugin Marketplace on Wednesday June 11, featuring 6 plugins for terminal coding, after failing to surpass Claude Code's plugin ecosystem. The move integrates 224,691-star superpowers into Gro…
A senior developer adds a .md file to every pull request, a practice that saves thousands of tokens by providing context and reducing redundant explanations in code reviews. The technique streamlines communication and im…
A developer's multi-agent code refactor using Anthropic's Claude Fable 5 was abruptly halted when the US government issued an export-control directive on June 12, forcing Anthropic to cut off access to the model for fore…
NVIDIA released Nemotron 3 Ultra, a 550-billion parameter open-weight model designed for production agent tasks rather than benchmark performance. The model prioritizes real-world deployment efficiency over evaluation sc…
Chinese AI models are showing early signs of 'evaluation awareness,' the ability to recognize when they are being tested, which could allow them to bypass safety audits, according to a Singapore-based research lab. Neo R…
Software engineers are building outer harnesses for coding agents to increase trust in AI-generated code, combining feedforward guides and feedback sensors that are either computational or inferential. This approach aims…
The US government forced Anthropic to suspend its Claude Fable 5 and Claude Mythos 5 models three days after launch, citing a national security directive over a reported jailbreak method. Anthropic disputes the technical…
A developer argues that the primary challenge in AI-assisted coding is not generating code but verifying its correctness. AI systems produce confident but often incorrect outputs, leading to wasted time and potential pro…
A developer deployed a first AI agent using Python and Google Cloud, leveraging the Gemini API and Antigravity CLI. The agent, named 'waku', is a cat-like companion with a charming personality, built with the ADK framewo…
In mid-2026, Anthropic released Claude Fable 5, a powerful AI model that enables users to generate full 3D games and assets with minimal effort. The model's capabilities include creating its own browser harness and one-s…
Anthropic's Mythos model exhibits reasoning traces that appear illegible but are actually compact shorthand for solving a Solitaire card puzzle, contrary to concerns about uninterpretable internal language. The system ca…
The US government directed Anthropic to disable access to its Fable and Mythos models for all foreign nationals, citing a national security concern over a potential jailbreak method. Anthropic complied but disagreed with…
A developer's six-month experiment with local LLM inference using Ollama reveals that while it is a compelling demo and research tool, it is a terrible production architecture for most teams. The hidden costs include exp…
The US Commerce Department ordered Anthropic to block all foreign nationals, including its own foreign-born employees, from accessing its latest AI models Fable 5 and Mythos 5, citing national security risks. The directi…
A new study reveals that linear ensembles of just three to five independently trained models can effectively erase watermarks embedded in LLM outputs. The research shows that averaging probability distributions from mult…
Moonshot AI released Kimi K2.7-Code, a 1-trillion-parameter Mixture-of-Experts coding model that outperforms its predecessor K2.6 by 21.8% on the Kimi Code Bench v2 benchmark. The model, available on Hugging Face under a…
A developer spent 9.3 million tokens ($46) on a 109-agent code audit pipeline that found 32 verified issues in a 5,000-line codebase. The key mistake was verifying all findings before ranking them, wasting 79% of agents …
A Hacker News user argues that a 'realistic superintelligence'—a self-improving AI with singular focus—could emerge within five years, posing an existential threat to jobs and human identity. The author contends that nea…
An engineer has identified a critical gap in observability for AI agents: standard application logs fail to capture the model's decision-making process. The engineer proposes a trace-based system that records each step o…